Research On Extreme Learning Machine Under The Cloud Environment

Extreme learning machine is a fast learning method to solve the single hidden layer neural network. It only needs to set the number of hidden layer nodes, randomly generates the input weights and hidden layer offsets which don’t need to do adjustment in the process, finally solves the minimum norm least squares problem. Therefore, extreme learning machine has the advantages of less training parameters, fast speed and good generalization performance.Although extreme learning machine has many advantages, as with the traditional machine learning algorithms, it has a real problem. That is to say, in the era of highly developed Internet, the volume of data grows exponentially, the traditional machine learning algorithms cannot effectively deal with large scale data, due to the limitation of machine memory. To solve this problem, this paper combines extreme learning machine with the present popular cloud computing platform, which uses the massively parallel processing systems to offer extreme machine learning algorithm the computation and storage space. That enables extreme machine learning algorithm to achieve the purpose of processing large scale data efficiently. The main content of this paper is as follows:Proposed an algorithm named extreme learning machine for classification based on Hadoop. According to the basic steps of extreme learning machine, firstly determine the number of hidden layer nodes and randomly generate the input weights and hidden layer offset, then calculate the hidden layer output matrix, finally get the output weights through the hidden layer output matrix and the samples. A Map Reduce job realizes a corresponding step, the output of previous job is the input of the next job, and so on. It takes advantage of the Hadoop’s parallel processing performance, and solves the inefficiency and memory exhaustion of traditional extreme learning machine in face of large scale data.Proposed an improved algorithm based on the above algorithm. Considering the computing time of each map and reduce, communication and networking time between map and reduce in a job, a Map Reduce job is used to realize all the steps of extreme learning machine(not including the preprocessing), it reduces the number of Map Reduce job, thereby reducing the data processing time and improving the algorithm speed.In conclusion, the main job in this paper is to put forward an algorithm called extreme learning machine for classification based on Hadoop and its improved algorithm, and prove the feasibility and effectiveness of the proposed algorithms.